Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: Parallel fiber PMD


I've included some feedback below.

Tom Palkert wrote:
> Even after reading all the 3 PMD emails I am still a believer in the
> parallel fiber PMD. I would like to throw out my rationale and look for
> feedback please.
> Note: I am not proposing that we replace any of the exising PMDs with
> the parallel fiber. They all have distinct application spaces. i.e. CWDM
> is required for 300m over existing fiber, serial for long distances etc.
> I just believe the very short reach application space justifies the cost
> savings and time to market provided by the parallel fiber.
> Technical feasibility:
> The technology to make a short wavelength 2.5 Gbps VCSEL is very
> similiar to the 1.25 Gbps VCSELs giving them, I believe, the lowest cost
> basis of any of the PMDs proposed.

2.5 Gbps is close to the proposed line rate for each of 4 lasers for a WAN PHY.
The line rate proposed for a LAN PHY is 3.125 Gbps. These two LAN/WAN rates are
exactly the same for parallel and WDM approaches and are also the same
regardless of laser wavelength. I agree with your lowest cost basis assumption
based on the same electronics as WDM but no need for multi-wavelength lasers,
filters and optical mux/demux devices. However, the required ribbon fiber and
ribbon connector raises total link cost. The question then becomes: At what
distance is a parallel optics link more economically feasible than a WDM or
Serial link?

> There are no issues with Open Fiber control. (The power is on each individual fiber not combined on a single fiber) i.e. Lower complexity.

On the contrary, Open Fiber Control is a real issue since a parallel fibre is
essentially treated as four bundled fibers drive by four lasers. The optical
power is 4X. However, this issue may be mitigated by implementing Open Fibre
Control by proposals such as:
presented by Mr. Ken Herrity of Blaze Networks in Albuquerque.
> 2.5 Gbps SERDES can be designed in std cmos technology. This makes the
> SERDES low cost or integrated with the framer.

Same goes for 3.125 Gbps SerDes. Actually, multiple integrated SerDes/CDR may be
significantly easier with 3.125 Gbps using 8B/10B code than with a 2.5 Gbps pure
scrambled code. Components that do the former are already sampling. I know of no
current components that do the latter. 
> The SERDES parts can be identical for either the WAN Phy or the LAN Phy
> assuming the OIF 622 LVDS interface is used.

You're lost me here. I don't know of any proposals and cannot conceive of any 10
GbE parallel fiber architecture employing 2.5 Gbps VCSELs which would need to
use an OIF 622 LVDS interface. Where would this interface be for either the WAN

> Other standards compatibility:
> OIF: They are generating interoperability agreements for Very Short
> Reach OC192 connections based on parallel fiber. The WAN Phy application
> will need to connect to the OC192 equipment. I think it would be highly
> desirable to have the OIF 4 wide parallel fiber solution specified in
> Infiniband: They are sending 10 Gbps data across very short links using
> parallel 2.5 Gbps VCSEL based fiber links. The optical modules should be
> interchangeable.

The above two paragraphs describe significantly disparate applications. The OIF
VSR OC-192 application transports SONET frames. InfiniBand transports packetized
data. InfiniBand is very similar to a 10 Gigabit Ethernet LAN application.
Furthermore, retiming and jitter specifications between SONET and InfiniBand/LAN
applications may be significantly different in order to address the cost
sensitivity differences between the two. I would say "It would be nice to have
optical modules that are interchangeable".
> General Market trends:
> Terabit routers and high density digital crossconnect applications are
> using large volumes of parallel fiber between equipment racks in their
> systems.  I would assume that they have done extensive cost analysis
> before committing to these programs. Most of the link analysis done for
> these applications assumes distances between 100 and 300m. Therefore, if
> a number of large volume users have analyzed a link that is virtually
> identical to the 10 GBE application and decided that 2.5 Gbps parallel
> fiber is the best solution, why shouldn't the 802.3ae group follow?
> It appears that 2.5 Gbps VCSELs will be readily available from multiple
> vendors either now or in the next couple of months.
> 2.5 Gbps cmos SERDES and integrated core macros will be readily
> available from multiple vendors either now or in the next couple of
> months.

I know that the above is true for 3.125 Gbps CMOS SerDes and integrated core
macros since this is already the case. I believe that multiple VCSEL vendors are
saying the same about their 3.125 Gbps capable VCSELs.
> Comments:

Note that the T11 10 Gigabit Fibre Channel (10 GFC) project has voted to specify
a parallel fiber physical variant (PMD). This vote was taken on June 6, 2000 in
Boise, ID.


Best Regards,
Richard Taborek Sr.                 Phone: 408-845-6102       
Chief Technology Officer             Cell: 408-832-3957
nSerial Corporation                   Fax: 408-845-6114
2500-5 Augustine Dr.        mailto:rtaborek@xxxxxxxxxxx
Santa Clara, CA 95054